11:15-12:00

• Part 4: Advanced Topics (30 min)
  • Imputation
  • Selective phenotyping
  • Simulation/Recombination
  • Visualization
• Exercises (10 min)
• Questions (5 min)

• Causes
  • GBS - alignment
  • Quality
  • SNP - hets not called
  • Different platforms

• But what if the reference panel is incomplete?

• Higher coverage
• Different platform
• More replicates
• …
• But sometimes that's not possible

  load('MissingData.RData')
  table(apply(missdat$founders, 2, function(x) sum(is.na(x))))

## 
##  0  1  2  3  4  5 
## 15 29 31 19  5  2

  nmiss <- apply(missdat$finals, 1, function(x) sum(is.na(x))/length(x))

  hist(nmiss, breaks=20, col="tomato")

• mpimpute
  • options for founders, finals, or both
  • relies on founder probability calculation

  impdat <- mpimpute(missdat)

## [1] "No chromosomes specified, will default to all"
## Using map groupings for groups. Remove map object if you want to regroup.
##  --Read the following data:
##   200  individuals
##   101  markers
##   2  phenotypes

  table(apply(impdat$founders, 2, function(x) sum(!is.na(x)))) 

## 
##   8 
## 101

  nmissi <- apply(impdat$finals, 1, function(x) sum(is.na(x))/length(x))
  sum(nmissi>0)

## [1] 0

  sum(is.na(impdat$founders)) 

## [1] 0

  sum(impdat$founders!=dat$founders)

## [1] 0

  sum(is.na(impdat$finals))/prod(dim(impdat$finals))

## [1] 0

  sum(impdat$finals!=dat$finals, na.rm=T)/sum(is.na(missdat$finals))

## [1] 0.08916084

• May want to test on your known data
  • Mask out some percentage, try imputation and estimate accuracy
• Affected by
  • marker density/spacing
  • sample size
  • type of genotyping platform
  • level of heterozygosity, etc.

• Phenotyping has overtaken genotyping in cost
  • done many times
  • at multiple scales
  • in multiple environments
  • for multiple traits
  • …
• So how do we best select a sample for phenotyping from a large set of genotyped individuals?

• A selection method which
  • is general, flexible and robust
  • best captures the genetic information of the population
  • maximized diversity, avoids genetic duplication
• Previous options (Jin et al. 2004, Jannink 2005) focused on specific designs, could not handle missing data

• Step 1: Compute pairwise distances between all individuals
• Step 2: Cluster distances into k groups
• Step 3: Select representative from each cluster

• Multiple stages of selection to refine QTL position
• Genomewide -> candidate gene level
• Selected lines will
  • have higher genetic diversity, so
  • are more likely to have recombination
  • and better resolve QTL location

• R package spclust
• Functions to
  • compute distance (spdist)
  • select lines (spclust, single- or multi-stage)
  • visualize (plot.spclust)

 library(spclust)
 load('SimulatedSP.RData')
 selLines <- spclust(dat, nlines=20, method="average")

## [1] "No chromosomes specified, will default to all"
## Using map groupings for groups. Remove map object if you want to regroup.
##  --Read the following data:
##   200  individuals
##   255  markers
##   2  phenotypes
## No required lines input; will only select a single-stage sample

  plot(selLines, type=2)  

• General MAGIC simulation
  • varying map
  • varying pedigree
  • varying QTL
• Missing data

• Assess imputation quality in your data
• Compare different designs
  • number of generations of advanced intercrossing
  • DH vs. RIL
  • MAGIC vs. NAM
• Test power for different approaches
• Generate empirical significance thresholds
• Estimate power for your map/data/founders
• See how theory compares to reality

• Generate different pedigrees
• Generate data from them
• Compare number of recombinations, size of haplotype blocks

 ped4 <- sim.mpped(4, 3, 200) # MAGIC4RIL 
 ped8 <- sim.mpped(8, 30, 200) # MAGIC8RIL, 30 funnels
 ped8ai2 <- sim.mpped(8, 1, 200, iripgen=2) # MAGIC8AI2RIL
 ped26nam <- generateNAMpedigree(26, 100) #NAMRIL

• AlphaMPSim (Hickey et al. 2014)
• Written in Fortran/R
• Faster and more memory efficient than using sim.mpcross
• For very large-scale simulations (>30K markers)

• Need to set up larger scale scripts
• Generate multiple datasets
  • Could keep observed founder genotypes, map, new progeny
  • Could keep observed progeny genotypes as well - new phenotype
  • Depends on how generalizable results need to be
• Analyze each dataset as you would observed data
• How often are QTL of different sizes detected?
• See Kover et al. (2009) for more details on procedure

load('datfinalPart2.RData')
## Based just on highest probability allele
nrecEst <- lapply(mppEst$estfnd, function(x) 
  apply(x, 1, function(y) return(sum(diff(y[!is.na(y)])!=0))))
mean(rowSums(do.call("cbind", nrecEst)))

## [1] 12.44

## Errors in the map can cause additional recombination events
load('Part2.RData')
nrecTrue <- lapply(mppTrue$estfnd, function(x) 
  apply(x, 1, function(y) return(sum(diff(y[!is.na(y)])!=0))))
mean(rowSums(do.call("cbind", nrecTrue)))

## [1] 11.954

## Based on forward-backward algorithm with penalty
source('nrec.R')
mean(nrec(mppEst, penalty=0)$totrec)

## [1] 13.797

mean(nrec(mppEst, penalty=1)$totrec)

## [1] 12.055

mean(nrec(mppEst, penalty=2)$totrec)

## [1] 10.941

nr <- nrec(mppEst, penalty=1)
boxplot(do.call("cbind", nr$nrec), col="tomato")

mppEst$pheno$nrec <- nrec(mppEst, penalty=1)$totrec
mprec <- mpIM(object=mppEst, responsename="nrec", ncov=0)
## No QTL found - but possible in real data

• ggplot2
• lattice
• RCircos
• Heatplus
• LDheatmap
• …

• Works on OS/X and Windows
• Uses:
  • Grouping markers, combining groups
  • Removing markers
  • Flipping blocks of markers
  • Re-ordering chunks of markers
• Note: currently only works for mpcross objects, but could relatively easily be extended to more general crosses

• plot.mpcross
  • Linkage map
  • Histogram of phenotype(s)
  • RF/LOD heatmap
• plot.mpprob
  • Percent of chromosome inherited from each founder
  • Haplotype mosaics for each chromosome
  • Founder inheritance across genome
• plot.mpqtl
  • QTL profile
  • Support interval

 load('datfinalPart2.RData')
 plot(datfinal)

mpp <- mpprob(datfinal, program="qtl")
plot(mpp)

• mapcomp function
  • subsets down to common markers/chromosomes
  • compares positions of markers with the same names
  • identifies markers with conflicting chromosomes
• summary.mapcomp
  • summarizes number of markers in each map and in common
  • identifies duplicated markers in each map
  • correlations between chromosomes in each map
• plot.mapcomp
  • scatterplot of markers positions for the two maps

load('Part2.RData')
mc <- mapcomp(dat, datfinal)
summary(mc)

## Number of markers in map1 is  505 
## Number of markers in map2 is  505 
## Number of common markers is  505 
## Number of duplicated markers in map1 is  0 
## Number of duplicated markers in map2 is  0 
## Number of markers with differing chromosomes between maps is  0 
## Correlations between chromosomes are: 
## 0.999142 0.991306 -0.999554 -0.9988986 0.9987963

plot(mc)

library(RCircos)
source('CircularIntx.R')
pmatrix <- matrix(runif(505*505, 0, 1), nrow=505)
plotCircIntx(pmatrix, threshold=1e-4, map=datfinal$map, file="CircEx.png")

Very basic simulation - generate 10 datasets each from 4-parent and 8-parent with these parameters - same map, 5 chromosomes, 51 markers per chromosome, 100 cM, evenly spaced - QTL + Chr 1, 47 cM, size=.5 + Chr 2, 13 cM, size=.3 + Chr 3, 81 cM, size=.1 - How often is each QTL detected?

For one of the 4-parent and 8-parent datasets generated for the simulation, what's the average number of recombinations?

Plot a histogram of the recombinations for each dataset.

Jin et al. 2004, Selective phenotyping for increased efficiency in genetic mapping studies. Genetics 168:2285-2293. doi: 10.1534/genetics.104.027524

Jannink 2005, Selective phenotyping to accurately map quantitative trait loci. Crop Science 45:901-908. doi: 10.2135/cropsci2004.0278

Hickey et al. 2014, AlphaMPSIM: flexible simulation of multi-parent crosses. Bioinformatics 30:2686-2688. doi: 10.1093/bioinformatics/btu206

Huang et al. 2013, Selecting subsets of genotyped experimental populations for phenotyping to maximize genetic diversity. TAG 126: 379-388. doi: 10.1007/s00122-012-1986-4

Huang et al. 2014, Efficient imputation of missing markers in low-coverage genotyping-by-sequencing data from multiparental crosses. Genetics 197:401-404. doi: 10.1534/genetics.113.158014

Kover et al. 2009, A multiparent advanced generation inter-cross to fine-map quantitative traits in Arabidopsis thaliana. PLoS Genetics doi: 10.1371/journal.pgen.1000551